Everyone loves to shake a stick at the healthcare industry for being backward. Fact is there is no lack of technology or data in healthcare.
Biggest challenge for healthcare providers is to identify what questions to ask the data. My team has implemented over 75 enterprise data warehouse projects in US healthcare industry. At the annual Seattle Code Camp, we discussed some of the examples of how data is used in the healthcare industry for compliance reporting (BI) and predictive analytics.
These slides are from Seattle Code Camp 2016, shares technologies, concepts and ideas for data science in the US healthcare industry.
2. ABOUT THE SPEAKER
2
Gaurav Garg “GG” is a business savvy IT leader playing the role of
trusted advisor and product development manager.
• Leads a Healthcare consulting practice for a national IT consulting
firm (www.calance.com).
• Participated in product strategy team - identified High Value
Questions resulting in 150+ new products and features in $16B GE
Healthcare product portfolio.
• Played the role of IT Director at UCLA Health System, acting as an
interface between the Clinical Directors and the IT Department.
• Led the delivery of 8 products and 70+ Enterprise Data Warehouse
projects in the Healthcare industry.
• P&L responsibility managing application development up to
$45MM.
• Presented and won Phase I, Phase II grants from the NIH under SBIR
funding to build a product in partnership with UCLA Health System.
• As a thought leader, GG publishes white papers, appears in
speaking engagements and guides healthcare startups at Cambia
Grove.
Gaurav Garg (GG)
https://www.linkedin.com/in/gauravkinatus@gaurav_kinatus
5. WHY IS HEALTHCARE DATA MESSY?
5
Integrated
Clinical
Information
System
Financial
System
Lab
Anatomic
Pathology
(6 vendors)
EMPI
(5 vendors)
Dictation
(4 vendors)
Interface
Engines
(7 vendors)
Ambulatory
Specialty
(4 vendors)
Hospice
(5 vendors)
Advanced
Visualization
(5 vendors)
Budgeting
(3 vendors)
Labor and
Delivery
(7 vendors)
Time and
Attendance
(3 vendors)
Ambulatory
EMR
(Small
practice)
(18 vendors)
Anesthesia
(1 vendor)
PACS
(31 vendors)
Patient
Monitoring
System
(1 vendor)
Laboratory
Blood Bank
(9 vendors)
CT Ultrasound
Other
medical
Devices
Ventilators
Fetal
monitors
Core
Measures
Rapid
Response
PHR
integration
Ambulatory
Inpatient
Throughput
Tracking
Infection
Tracking
Quality
Scorecard
Care
Quality
Analysis
Charge &
Payment
Analysis
IHI
Triggers
6. BI & DATA WAREHOUSE EXAMPLES
6
Clinical and
Syndrome
Surveillance
MEWS; HAC Risk Scoring (DVT,
DU, CLBSI, CA-UTI, CTI, PU,
Wound); Sepsis; school
absenteeism; bio-surveillance
1 Patient 1
Chart
Optimizing
Resource
Utilization
Imaging Tracking; Room
Utilization; Capital
Investment Optimization;
Payer Compliance; Leakage;
LOS; Re-Admissions Manager
(RAM)
Improving
Physician
Satisfaction
Protocol
Compliance
Tracking
Pharmacy Protocol, Core
Measures (AMI, HF, PN, SCIP,
MU), Nurse Admission
Tracking
Improving
Patient
Satisfaction
Data explosion from EMR adaption has
created lots of data. Calance has
implemented over data warehouse
projects in over 70 hospitals and HIEs.
• EMPI Integration – we have experience
integrating with popular EMPI solutions and
troubleshooting performance issues.
• Enterprise Interface Engine – implementation,
upgrade, migrate, configure interfaces.
Caradigm Intelligence Platform (aka Amalga),
Orion Rhapsody, eGate, Intersystem
Ensemble, BizTalk, SSIS.
• Data Warehouse/Data Lake – we have
implemented and managed some of the
largest data warehouse/data lake
infrastructure in the healthcare industry
(largest @ 3 petabyte data).
• Scorecard/Reports – data visualization for
scorecards, dashboards and reports on
SharePoint, Roambi, tableau.
• Predictive Analytics – build risk stratification
and other predictive analytics algorithms
using hadoop ecosystem and RStudio.
7. 7
DATA PIPELINE
EMPI
Staging Enterprise
Service Bus
Data
Warehouse
BI Analytics,
Dashboards,
Visualization
& Report
Coordination
Browser
Mobile Device
Redshift,
NoSQL,
SQL Server
Authorization and DevOpsData Governance
Products and Logos acknowledged to respective owners. Logos used to for illus
12. 12
PREDICTIVE ANALYTICS EXAMPLE –AUTOBED
https://www.youtube.com/watch?v=eI1l_s4zo_s
Predictive Analytics Example –AutoBed
Watch the video on the following Slide
14. Know Your Patients Clinical ProtocolsBusiness Intelligence & Predictive
Analytics
Cohort Identification Standardized Care PathPatient Demographics
Lifestyle
Medical History
Risk Stratification
Adverse Event Prediction
Family History
Proactive Appointments
Treatment Effectiveness
Patient Generated Data
Variance Management
Financial Review
Patterns
Treatment Adjustment
Actionable Intelligence
POPULATION HEALTH EXPLAINED
15. PROTOCOL COMPLIANCE
Clinical desktop integration using
“Knowledge Hub”
Protocol Definition and Matching
Clinician view, compliance officer view,
reporting moduleLanguage & Terminologies
Data Aggregation
TouchPoint360
Data existing in hospital systems
Calance Protocol Compliance Framework
20. 20
AMBULATORY
• Is my practice setup to be successful
as a Pay-for-Performance contract or
should I stay as pay per service?
• Should I go an form an Accountable
Care Organization (ACO) with 75
other physicians or join an Integrated
Delivery Network (IDN)?
• What are my compliance
requirements with different
contracts?
• How do I rapidly understand the
business implications of (program de
jour) for my specific practice so I can
rapidly make informed decisions to
participate (or not)?
21. 21
HIGH ACUITY CARE
• What is the probability of patient
developing an infection?
• Do I have everything for tomorrow?
• How are we doing on infection
management (SEPSIS)? Is there
anything we can do to use the
devices to proactive alerts?
• What is the correlation between cost
index and length of stay?
22. 22
DIGITAL PATHOLOGY
• Find similar cases
• What other information do you need
for diagnosis? Additional images,
molecular image, additional test
results, additional symptoms
• How to close the loop between
Radiologist and Pathologist to reduce
discordance?
23. 23
RADIOLOGY
• Automatically detect abnormalities in
the image
• Which Radiologist results in
good/bad follow-up actions?
• What are my compliance
requirements with different
contracts?
• Is this the right procedure, modality
and protocol?
24. 24
REVENUE CYCLE
• Recommend an immediate step to avoid
hospitalization. e.g. cab for picking up Rx
• Who is the best provider to treat this
patient?
• What is the best site for this patient?
• Coding accuracy dashboard
Editor's Notes
Healthcare data is messy –
Wide variety of data.
Highly specialized software to automate clinical workflows by specialty/department.
No two implementations of the same department workflows are same.
Too many versions of “standards” for data exchange. HL7 has 30+ variations and none of them are backward compatible.
Implementation of interface at every hospital, even within a hospital system, is likely to be different. Think “Misc” field in database; everyone can change the data and structure within a HL7.
Unstructured data aka clinical notes.
Products from the same vendor does not talk to each other.
(Only in the Healthcare industry), the data is owned by the system where it is generated. This means, the developer needs consent and cooperation from the vendor to integrate with their system.
Data driven initiatives in the healthcare industry are not new. There is no shortage of technology and data in healthcare.
Traditional retrospective BI focused on compliance and quality reports.
Large hospitals have played with data integration in some form or other for over a decade.
Microsoft’s HealthVault, Google Health were exampled of failed attempts at giving patient access to their own health records. Latest in that category is Apple’s HealthKit on iOS devices. Apple HealthKit now supports Continuity of Care Record (CCR) format.
HealthVault integrated with home health devices before wearable was a thing – blood glucose meter, pulse oxymeters (for asthma patients), heart rate monitors etc.
Some technologies you may need to learn if you want to play with healthcare data.
Calance is a technology agnostic systems integration firm. We do not endorse any specific product.
Tools for retrospective BI are different from predictive analytics.
EMPI – Electronic Master Patient Index is widely used to identify and merge patient records from multiple sources into a comprehensive electronic record.
Interface engines like Orion Rhapsody and Intersystems’ Ensemble are widely used to deal with different formats and standards. There are open source tools like Talend, Mirth Connect and others that can be used for healthcare interfaces.
Columnar databases are better at data warehousing – creating wide views, slicing/dicing massive amounts of data. Focus less on ACID functionality and more on aggregate data performance.
Transactional databases are better since they focus on ACID (Atomicity, Consistency, Isolation, Durability) properties.
Traditional database management systems e.g. SQL Server, Oracle etc. were hybrid databases attempting to address the needs of both the Transactional systems and the data warehouse. Hybrid databases spend too much administration time on ACID functions and administration, making them bad at large dataset manipulation. Since these DBMS are geared towards structured data, hence the star schemas.
Microsoft and Oracle has introduced new columnar databases specifically for data warehousing.
Unstructured data e.g. clinical notes is stored differently and manipulated differently.
Think about you will manage data quality.
If you plan to work with healthcare industry, get used to the cash flow. This video explains the relationship between patient (employees), employers, providers (hospital/clinicians) and the payers (insurance companies).
Because hospitals have so many different moving parts, the cause and effects are rarely in the same department.Here is an example of predictive analytics. In this video, we show how predictive analytics and be used to find the root cause of a problem in surgical ward.
In the US, there is new push for Accountable Care Organization (ACO) and Affordable Care Act (Obama Care). Hospitals are under huge pressure to change their business model. This slide summarizes what is happening as a results of these new initiatives. Insurance companies are forcing the hospitals from move from a Time and Materials business model to a Fixed Fee business models.
aka – “Pay for service” to “Pay for performance” reimbursement model.
What does Population Health Management means?
CRM + Predictive Analytics + Standardized Care paths - variance analysis = actionable intelligence and proactive care
Example of Calance product – Clinical Protocol Compliance. This product was showcases at HIMSS 2013 at Caradigm booth.
Predictive analytics based physician dashboard to show “is there anything, I need to do before I go home”.
Protocol level details and patients who need my (clinicians) attention right now.
Retrospective dashboard showing variance at aggregate level. Which physicians are using agreed protocols and relatively more efficient?
Some more ideas for predictive analytics in the subsequent slides.
I volunteer office hours at Cambia Grove regularly to help Seattle based healthcare startups. Anyone can book a slot to bounce ideas (for free). Feel free to connect on LinkedIn, @gaurav_kinatus or through Cambia Grove https://www.cambiagrove.com/calendar/all.